Merge remote-tracking branch 'upstream/sycl' into sycl-fix-group-algorithm

Author: dklochkov-emb

llvm/lib/SYCLNativeCPUUtils/CMakeLists.txt

@@ -34,15 +34,15 @@ endif()
 if(NATIVECPU_USE_OCK)
   if(NATIVECPU_OCK_USE_FETCHCONTENT)
     set(OCK_GIT_INTERNAL_REPO "https://github.com/uxlfoundation/oneapi-construction-kit.git")
-    # commit ffef31717750d3f15bfd2f18d9cd7f4677fe9d3e
-    # Merge: 574307afde 0b00238554
+    # commit 03e6497dfcc06b0ef6bdeeb8fefd7fdc24de4287
+    # Merge: df8de76a56 413b59ebe2
     # Author: Harald van Dijk <[email protected]>
-    # Date:   Fri Apr 4 16:26:38 2025 +0100
+    # Date:   Sat May 17 20:52:27 2025 +0100
     # 
-    #     Merge pull request #751 from hvdijk/llvm21-address-space
+    #     Merge pull request #822 from coldav/colin/remove_ca_enable_api
     #     
-    #     [LLVM 21] Take address space into account for legality.
-    set(OCK_GIT_INTERNAL_TAG ffef31717750d3f15bfd2f18d9cd7f4677fe9d3e)
+    #     Remove CA_ENABLE_API as a CMake option
+    set(OCK_GIT_INTERNAL_TAG 03e6497dfcc06b0ef6bdeeb8fefd7fdc24de4287)
 
     # Overwrite OCK_GIT_INTERNAL_REPO/OCK_GIT_INTERNAL_TAG if the corresponding options are set
     if(OCK_GIT_REPO)

sycl/doc/extensions/experimental/sycl_ext_oneapi_bindless_images.asciidoc

@@ -43,14 +43,12 @@ change incompatibly in future versions of {dpcpp} without prior notice.
 *Shipping software products should not rely on APIs defined in this
 specification.*
 
-
 == Backend support status
 
-This extension is currently implemented in {dpcpp} only for GPU devices and
-only when using the CUDA backend.  Attempting to use this extension in
-kernels that run on other devices or backends will not work. 
-Be aware that the compiler may not be able to issue a diagnostic to
-warn you if this happens.
+This extension is currently implemented in {dpcpp} only for GPU devices. CUDA,
+HIP, and LevelZero backends are supported. Not all image and interoperability
+features may be supported on all backends. We provide device aspects which users
+can query to check support for specific features.
 
 == Overview
 
@@ -2481,13 +2479,6 @@ include::../../../test-e2e/bindless_images/examples/example_5_sample_cubemap.cpp
 include::../../../test-e2e/bindless_images/examples/example_6_import_memory_and_semaphores.cpp[lines=14..-1]
 ```
 
-== Implementation notes
-
-The current DPC++ prototype only implements the proposal for the CUDA backend,
-however we are actively exploring Level Zero with SPIR-V.
-We are looking at other backend as well in order to ensure the extension can 
-work across different backends.
-
 == Issues
 
 === No dependency tracking
@@ -2515,12 +2506,6 @@ The ability to create an image with 3 channels depends on the backend.
 There is currently no way to query a backend whether it supports this feature.
 This query should be added in a later revision of the proposal.
 
-=== Not supported yet
-
-These features still need to be handled:
-
-* Level Zero and SPIR-V support
-
 == Revision History
 
 [frame="none",options="header"]
@@ -2740,3 +2725,6 @@ These features still need to be handled:
 |6.8|2025-03-13| - Add support for importing timeline semaphores.
 |6.9|2025-03-18| - Add new `gather` image type and the accompanying
                    `gather_image` function.
+|6.10|2025-05-09| - Add `unmap_external_image_memory` and 
+                    `unmap_external_linear_memory` APIs.
+                  - Clarify how and when external memory should be unmapped.

sycl/include/sycl/ext/khr/free_function_commands.hpp renamed to sycl/include/sycl/khr/free_function_commands.hpp

@@ -74,7 +74,6 @@
 #include <sycl/ext/intel/experimental/task_sequence_properties.hpp>
 #include <sycl/ext/intel/experimental/usm_properties.hpp>
 #include <sycl/ext/intel/usm_pointers.hpp>
-#include <sycl/ext/khr/free_function_commands.hpp>
 #include <sycl/ext/oneapi/bfloat16.hpp>
 #include <sycl/ext/oneapi/bindless_images.hpp>
 #include <sycl/ext/oneapi/device_global/device_global.hpp>
@@ -123,4 +122,5 @@
 #include <sycl/ext/oneapi/virtual_mem/physical_mem.hpp>
 #include <sycl/ext/oneapi/virtual_mem/virtual_mem.hpp>
 #include <sycl/ext/oneapi/weak_object.hpp>
+#include <sycl/khr/free_function_commands.hpp>
 #include <sycl/khr/group_interface.hpp>

sycl/include/sycl/sycl.hpp

@@ -91,8 +91,15 @@ function(add_sycl_rt_library LIB_NAME LIB_OBJ_NAME)
 
   # Feature-specific compilation and link step setup
 
+  # Unlike for sycl library, for LLVMSupport we have only one version for a given build,
+  # so, we link LLVMSupport lib to matching sycl version only.
   if (SYCL_ENABLE_STACK_PRINTING)
-    target_compile_definitions(${LIB_OBJ_NAME} PUBLIC ENABLE_STACK_TRACE)
+    if(NOT MSVC OR (CMAKE_BUILD_TYPE STREQUAL "Debug" AND ARG_COMPILE_OPTIONS MATCHES ".*MDd.*") OR
+      (NOT CMAKE_BUILD_TYPE STREQUAL "Debug" AND NOT ARG_COMPILE_OPTIONS MATCHES ".*MDd.*"))
+        add_dependencies(${LIB_NAME} LLVMSupport)
+        target_compile_definitions(${LIB_OBJ_NAME} PUBLIC ENABLE_STACK_TRACE)
+        target_link_libraries(${LIB_NAME} PRIVATE LLVMSupport)
+    endif()
   endif()
 
   # TODO: Enabled for MSVC
@@ -186,11 +193,6 @@ function(add_sycl_rt_library LIB_NAME LIB_OBJ_NAME)
     endif()
   endif()
 
-  # Link with LLVMSupport for shared utilities.
-  add_dependencies(${LIB_NAME} LLVMSupport)
-  target_link_libraries(${LIB_NAME} PRIVATE LLVMSupport)
-  target_include_directories(${LIB_OBJ_NAME} SYSTEM PRIVATE ${LLVM_MAIN_INCLUDE_DIR})
-
   check_linker_flag(CXX "-Wl,--gc-sections" LINKER_SUPPORTS_WL_GC_SECTIONS)
   if(LINKER_SUPPORTS_WL_GC_SECTIONS)
     # Reduces the size of the resulting library by having the linker perform

sycl/source/CMakeLists.txt

@@ -35,6 +35,45 @@
 #define __SYCL_DEVICE_BINARY_TARGET_LLVM_NVPTX64 "llvm_nvptx64"
 #define __SYCL_DEVICE_BINARY_TARGET_LLVM_AMDGCN "llvm_amdgcn"
 
+/// Device binary image property set names recognized by the SYCL runtime.
+/// Name must be consistent with
+/// PropertySetRegistry::SYCL_SPECIALIZATION_CONSTANTS defined in
+/// PropertySetIO.h
+#define __SYCL_PROPERTY_SET_SPEC_CONST_MAP "SYCL/specialization constants"
+/// PropertySetRegistry::SYCL_SPEC_CONSTANTS_DEFAULT_VALUES defined in
+/// PropertySetIO.h
+#define __SYCL_PROPERTY_SET_SPEC_CONST_DEFAULT_VALUES_MAP                      \
+  "SYCL/specialization constants default values"
+/// TODO: remove req mask when sycl devicelib online linking path is removed.
+/// PropertySetRegistry::SYCL_DEVICELIB_REQ_MASK defined in PropertySetIO.h
+#define __SYCL_PROPERTY_SET_DEVICELIB_REQ_MASK "SYCL/devicelib req mask"
+/// PropertySetRegistry::SYCL_DEVICELIB_METADATA defined in PropertySetIO.h
+#define __SYCL_PROPERTY_SET_DEVICELIB_METADATA "SYCL/devicelib metadata"
+/// PropertySetRegistry::SYCL_KERNEL_PARAM_OPT_INFO defined in PropertySetIO.h
+#define __SYCL_PROPERTY_SET_KERNEL_PARAM_OPT_INFO "SYCL/kernel param opt"
+/// PropertySetRegistry::SYCL_KERNEL_PROGRAM_METADATA defined in PropertySetIO.h
+#define __SYCL_PROPERTY_SET_PROGRAM_METADATA "SYCL/program metadata"
+/// PropertySetRegistry::SYCL_MISC_PROP defined in PropertySetIO.h
+#define __SYCL_PROPERTY_SET_SYCL_MISC_PROP "SYCL/misc properties"
+/// PropertySetRegistry::SYCL_ASSERT_USED defined in PropertySetIO.h
+#define __SYCL_PROPERTY_SET_SYCL_ASSERT_USED "SYCL/assert used"
+/// PropertySetRegistry::SYCL_EXPORTED_SYMBOLS defined in PropertySetIO.h
+#define __SYCL_PROPERTY_SET_SYCL_EXPORTED_SYMBOLS "SYCL/exported symbols"
+/// PropertySetRegistry::SYCL_IMPORTED_SYMBOLS defined in PropertySetIO.h
+#define __SYCL_PROPERTY_SET_SYCL_IMPORTED_SYMBOLS "SYCL/imported symbols"
+/// PropertySetRegistry::SYCL_DEVICE_GLOBALS defined in PropertySetIO.h
+#define __SYCL_PROPERTY_SET_SYCL_DEVICE_GLOBALS "SYCL/device globals"
+/// PropertySetRegistry::SYCL_DEVICE_REQUIREMENTS defined in PropertySetIO.h
+#define __SYCL_PROPERTY_SET_SYCL_DEVICE_REQUIREMENTS "SYCL/device requirements"
+/// PropertySetRegistry::SYCL_HOST_PIPES defined in PropertySetIO.h
+#define __SYCL_PROPERTY_SET_SYCL_HOST_PIPES "SYCL/host pipes"
+/// PropertySetRegistry::SYCL_VIRTUAL_FUNCTIONS defined in PropertySetIO.h
+#define __SYCL_PROPERTY_SET_SYCL_VIRTUAL_FUNCTIONS "SYCL/virtual functions"
+/// PropertySetRegistry::SYCL_IMPLICIT_LOCAL_ARG defined in PropertySetIO.h
+#define __SYCL_PROPERTY_SET_SYCL_IMPLICIT_LOCAL_ARG "SYCL/implicit local arg"
+/// PropertySetRegistry::SYCL_REGISTERED_KERNELS defined in PropertySetIO.h
+#define __SYCL_PROPERTY_SET_SYCL_REGISTERED_KERNELS "SYCL/registered kernels"
+
 /// Program metadata tags recognized by the PI backends. For kernels the tag
 /// must appear after the kernel name.
 #define __SYCL_PROGRAM_METADATA_TAG_REQD_WORK_GROUP_SIZE "@reqd_work_group_size"

sycl/source/detail/compiler.hpp

@@ -12,8 +12,6 @@
 // For device image compression.
 #include <detail/compression.hpp>
 
-#include <llvm/Support/PropertySetIO.h>
-
 #include <algorithm>
 #include <cstdlib>
 #include <cstring>
@@ -187,39 +185,28 @@ void RTDeviceBinaryImage::init(sycl_device_binary Bin) {
     // try to determine the format; may remain "NONE"
     Format = ur::getBinaryImageFormat(Bin->BinaryStart, getSize());
 
-  SpecConstIDMap.init(
-      Bin, llvm::util::PropertySetRegistry::SYCL_SPECIALIZATION_CONSTANTS);
+  SpecConstIDMap.init(Bin, __SYCL_PROPERTY_SET_SPEC_CONST_MAP);
   SpecConstDefaultValuesMap.init(
-      Bin, llvm::util::PropertySetRegistry::SYCL_SPEC_CONSTANTS_DEFAULT_VALUES);
-  DeviceLibReqMask.init(
-      Bin, llvm::util::PropertySetRegistry::SYCL_DEVICELIB_REQ_MASK);
-  DeviceLibMetadata.init(
-      Bin, llvm::util::PropertySetRegistry::SYCL_DEVICELIB_METADATA);
-  KernelParamOptInfo.init(
-      Bin, llvm::util::PropertySetRegistry::SYCL_KERNEL_PARAM_OPT_INFO);
-  AssertUsed.init(Bin, llvm::util::PropertySetRegistry::SYCL_ASSERT_USED);
-  ImplicitLocalArg.init(
-      Bin, llvm::util::PropertySetRegistry::SYCL_IMPLICIT_LOCAL_ARG);
-  ProgramMetadata.init(Bin,
-                       llvm::util::PropertySetRegistry::SYCL_PROGRAM_METADATA);
+      Bin, __SYCL_PROPERTY_SET_SPEC_CONST_DEFAULT_VALUES_MAP);
+  DeviceLibReqMask.init(Bin, __SYCL_PROPERTY_SET_DEVICELIB_REQ_MASK);
+  DeviceLibMetadata.init(Bin, __SYCL_PROPERTY_SET_DEVICELIB_METADATA);
+  KernelParamOptInfo.init(Bin, __SYCL_PROPERTY_SET_KERNEL_PARAM_OPT_INFO);
+  AssertUsed.init(Bin, __SYCL_PROPERTY_SET_SYCL_ASSERT_USED);
+  ImplicitLocalArg.init(Bin, __SYCL_PROPERTY_SET_SYCL_IMPLICIT_LOCAL_ARG);
+  ProgramMetadata.init(Bin, __SYCL_PROPERTY_SET_PROGRAM_METADATA);
   // Convert ProgramMetadata into the UR format
   for (const auto &Prop : ProgramMetadata) {
     ProgramMetadataUR.push_back(
         ur::mapDeviceBinaryPropertyToProgramMetadata(Prop));
   }
-  ExportedSymbols.init(Bin,
-                       llvm::util::PropertySetRegistry::SYCL_EXPORTED_SYMBOLS);
-  ImportedSymbols.init(Bin,
-                       llvm::util::PropertySetRegistry::SYCL_IMPORTED_SYMBOLS);
-  DeviceGlobals.init(Bin, llvm::util::PropertySetRegistry::SYCL_DEVICE_GLOBALS);
-  DeviceRequirements.init(
-      Bin, llvm::util::PropertySetRegistry::SYCL_DEVICE_REQUIREMENTS);
-  HostPipes.init(Bin, llvm::util::PropertySetRegistry::SYCL_HOST_PIPES);
-  VirtualFunctions.init(
-      Bin, llvm::util::PropertySetRegistry::SYCL_VIRTUAL_FUNCTIONS);
-  RegisteredKernels.init(
-      Bin, llvm::util::PropertySetRegistry::SYCL_REGISTERED_KERNELS);
-  Misc.init(Bin, llvm::util::PropertySetRegistry::SYCL_MISC_PROP);
+  ExportedSymbols.init(Bin, __SYCL_PROPERTY_SET_SYCL_EXPORTED_SYMBOLS);
+  ImportedSymbols.init(Bin, __SYCL_PROPERTY_SET_SYCL_IMPORTED_SYMBOLS);
+  DeviceGlobals.init(Bin, __SYCL_PROPERTY_SET_SYCL_DEVICE_GLOBALS);
+  DeviceRequirements.init(Bin, __SYCL_PROPERTY_SET_SYCL_DEVICE_REQUIREMENTS);
+  HostPipes.init(Bin, __SYCL_PROPERTY_SET_SYCL_HOST_PIPES);
+  VirtualFunctions.init(Bin, __SYCL_PROPERTY_SET_SYCL_VIRTUAL_FUNCTIONS);
+  RegisteredKernels.init(Bin, __SYCL_PROPERTY_SET_SYCL_REGISTERED_KERNELS);
+  Misc.init(Bin, __SYCL_PROPERTY_SET_SYCL_MISC_PROP);
 }
 
 std::atomic<uintptr_t> RTDeviceBinaryImage::ImageCounter = 1;

sycl/source/detail/device_binary_image.cpp

@@ -12,8 +12,6 @@
 #include <sycl/detail/os_util.hpp>
 #include <sycl/detail/ur.hpp>
 
-#include <llvm/Support/PropertySetIO.h>
-
 namespace sycl {
 inline namespace _V1 {
 namespace detail {
@@ -267,7 +265,7 @@ sycl_device_binaries jit_compiler::createDeviceBinaries(
 
     for (const auto &FPS : DevImgInfo.Properties) {
       bool IsDeviceGlobalsPropSet =
-          FPS.Name == llvm::util::PropertySetRegistry::SYCL_DEVICE_GLOBALS;
+          FPS.Name == __SYCL_PROPERTY_SET_SYCL_DEVICE_GLOBALS;
       PropertySetContainer PropSet{FPS.Name.c_str()};
       for (const auto &FPV : FPS.Values) {
         if (FPV.IsUIntValue) {

sycl/source/detail/jit_compiler.cpp

@@ -34,8 +34,6 @@
 
 #include <sycl/ext/oneapi/matrix/query-types.hpp>
 
-#include <llvm/Support/PropertySetIO.h>
-
 #include <algorithm>
 #include <cassert>
 #include <cstdint>
@@ -1865,8 +1863,7 @@ static bool isBfloat16DeviceLibImage(sycl_device_binary RawImg,
   for (ImgPS = RawImg->PropertySetsBegin; ImgPS != RawImg->PropertySetsEnd;
        ++ImgPS) {
     if (ImgPS->Name &&
-        !strcmp(llvm::util::PropertySetRegistry::SYCL_DEVICELIB_METADATA,
-                ImgPS->Name)) {
+        !strcmp(__SYCL_PROPERTY_SET_DEVICELIB_METADATA, ImgPS->Name)) {
       if (!LibVersion)
         return true;
 
@@ -1894,8 +1891,7 @@ getExportedSymbolPS(sycl_device_binary RawImg) {
   for (ImgPS = RawImg->PropertySetsBegin; ImgPS != RawImg->PropertySetsEnd;
        ++ImgPS) {
     if (ImgPS->Name &&
-        !strcmp(llvm::util::PropertySetRegistry::SYCL_EXPORTED_SYMBOLS,
-                ImgPS->Name))
+        !strcmp(__SYCL_PROPERTY_SET_SYCL_EXPORTED_SYMBOLS, ImgPS->Name))
       return ImgPS;
   }